Limit Order Backtests: Use Counterfactual Fill Ranges

Limit order backtests: Compare declared fill scenarios because a touched bar price alone cannot establish queue position.

Limit-order research should report counterfactual fill ranges when the available data cannot identify one execution. Choosing the word limit does not specify queue position, arrival latency, available quantity, or how an unfilled order changes later exposure. A plausible-looking fill is still a modeling assumption.

The Reddit discussion debated market versus limit execution assumptions. Its performance claims remain unverified, and some replies were inaccessible. This article develops an original offline scenario exercise rather than adopting a comment's execution claim.

Limit order backtests: separate price permission from execution

SEC order guidance explains limit-price constraints. For research, also state submission time, earliest arrival, expiry, quantity, fee basis, and what evidence permits a fill. A historical last trade at the limit may have occurred before arrival or consumed only a small quantity. Your hypothetical order was absent from the observed market.

Recorded depth can constrain a scenario without revealing exactly what would happen after an added order. A cancellation, priority rule, hidden order, or market response can change the outcome. Avoid calling an optimistic model a reconstruction merely because it uses more detailed data than daily bars.

A synthetic three-scenario fixture

Imagine an intended purchase of 100 shares at a $50 limit, followed by an observed sell opportunity at $50.20. Build three invented execution cases: all 100 fill, only 40 fill, and none fill. Before fees, the corresponding gains are $20, $8, and zero realized gain. The missed-fill case also retains its capital, which can change subsequent decisions.

Add a fourth case with a delayed fill after the apparent opportunity has passed. Its outcome must follow the later chronological prices, not receive the earlier favorable exit. Record filled quantity and remaining quantity independently. When expiry or cancellation occurs, preserve any late fill that the declared event sequence allows.

Do not calibrate from a paper label alone

Alpaca documents paper/live differences. That documentation describes one simulator, not every broker or research engine. Compare your own recorded intents, acknowledgments, fills, cancellations, and account states using a declared adapter contract. Cumulative quantity snapshots and incremental fills require different normalization; duplicate events must not create additional shares.

Keep execution stress separate from signal modification. If a pessimistic scenario also removes trades, shrinks positions, and changes entries, its net difference cannot be assigned solely to fills. Report each changed assumption and retain unchanged controls.

Turn uncertainty into a reviewable report

Show outcome ranges, unfilled demand, adverse post-fill price movement, fees, and terminal holdings for each scenario. A wide range is useful evidence that the available data supports weak identification. Do not replace it with the midpoint and call that expected performance without a separate calibration argument.

Find the offline execution workflow at Tradewink Advisory Research. Your first action is to create the 100-share fixture and verify quantity conservation in every case. Then insert observed timing and broker-specific event semantics where evidence exists. These exercises assess sensitivity; they do not guarantee liquidity or instruct live order placement.

Continue the research

Adaptive Backtest Overfitting: Log Research Decisions and Arbitrage Execution Risk: Audit the Unhedged Leg develop adjacent checks. Use a small synthetic fixture first; these educational diagnostics do not validate a trading edge or change account settings.